Randomized trial demonstrates improved photovoltaic power forecasting using adaptive clustering in real-time data, suggesting better solar energy management.
Key Points
This research aims to enhance photovoltaic power forecasting accuracy through a hybrid framework that incorporates clustering and deep learning techniques.
Integrates adaptive feature-weighted dynamic time warping (AFDTW) clustering with an improved temporal convolutional network–long short-term memory (ITCN–LSTM) model.
Employs an adaptive weighting mechanism based on Markov distance to prioritize high-variance features of photovoltaic output.
Uses a real dataset from a 1.8-MW PV plant in Australia for empirical evaluations.
The proposed method outperforms benchmark models across multiple forecasting horizons ranging from 5 minutes to 4 hours.
Significant improvements in power-forecasting accuracy were observed, supporting more effective utilization of solar energy resources.